The management of power consumption in large high power performance is critical to the overall performance of these systems. There are various techniques for reducing power consumption. One technique involves placing any unused servers in a sleep state. However, these unused servers may be connected to network switch ports that are still active and remain powered on. Network switch ports connected to unused servers may still actively auto-negotiate link connectivity, thus unnecessarily consuming power. Since large computing clusters may include thousands of nodes, large amounts of power may be wasted in such computing clusters. For at least these reasons, it is desired to provide improved techniques and systems for reducing server power consumption.